The Agent Skills Directory

[INDIRECT_PROMPT_INJECTION]: The skill instructs the agent to analyze external, potentially untrusted data which provides a surface for indirect prompt injection attacks.
Ingestion points: The agent is directed to use git log --oneline and git diff to review project history and code changes when determining version bumps (found in references/scenarios.md).
Boundary markers: There are no explicit delimiters or instructions provided to the agent to ignore or isolate instructions that might be embedded within commit messages or source code comments.
Capability inventory: The skill allows the agent to modify files (Write, Edit) and execute specific shell commands (git, npx changeset), which could be leveraged if the agent inadvertently follows instructions found in the git data.
Sanitization: No content validation, filtering, or sanitization logic is present to handle the data retrieved from the git repository.

creating-changesets